Remote Sensing Labs
Browse all docs

Sentinel-2

Sentinel-2 is the optical workhorse of open Earth observation, and most of the practical workflows on this site quietly assume it. Operated as part of Europe’s Copernicus program, it is a constellation of near-identical satellites sharing one orbit — nominally two operating at a time, half a revolution apart — so that between them they revisit the same ground far sooner than a single satellite could. That combination — a wide imaging swath, a rich set of visible, red-edge, near-infrared, and shortwave-infrared bands, and a frequent, free, systematically-acquired global record — is why Sentinel-2 shows up again and again in vegetation monitoring, agriculture, urban mapping, and water studies. What follows is about the mission as a data source: what its measurements are good for, what limits them, and how to reason about using them.

What Sentinel-2 contributes

The value of a data source is the questions it lets you answer well, and Sentinel-2’s design points squarely at monitoring the changing land surface. Two satellites sharing the orbit give a short effective revisit at the equator and shorter still toward the poles where swaths overlap, which means a given field or shoreline is imaged often enough to catch the rhythm of a growing season rather than a single lucky snapshot. The swath is wide, so regional and national coverage accumulates quickly, and the archive is consistent in design: the same instrument design, the same bands, and the same acquisition strategy over years, which is what a Time Series needs to be trustworthy — provided the processing changes described below are handled. It is optical and sun-dependent, so it sees reflected daylight and not through cloud, but within that envelope it is one of the most useful public records for watching how the surface looks and how that look changes.

Spatial, spectral, and temporal character

Sentinel-2’s usefulness comes from how its three kinds of resolution fit together, a balance the Resolution concept frames in general. Spatially, its bands are not all recorded at the same pixel size: the core visible and near-infrared bands are its sharpest, several red-edge and shortwave-infrared bands are somewhat coarser, and a few atmospheric bands are coarser still. That tiered design is deliberate — the bands that most need fine detail get it, while bands used for atmospheric correction or broad spectral context are collected at a scale matched to their purpose. In practice this means you choose bands with their native detail in mind, and resample deliberately rather than assuming every band shares one grid.

Spectrally, the sensor spans the visible through the shortwave-infrared with a band set chosen to read the land surface, and the Spectral Bands concept explains why each region matters. The visible and near-infrared bands carry the vegetation signature that drives NDVI and true- and false-color composites; the shortwave-infrared bands respond to moisture in leaves and soil and help separate snow, cloud, and burned ground; and a distinctive feature of Sentinel-2 is its multiple red-edge bands, sampling the steep rise between red and near-infrared where vegetation reflectance changes fastest. That red-edge sampling is part of why the mission is prized for crop and canopy work, because it resolves gradations of plant condition that a coarser red/NIR pair blurs together.

Temporally, the constellation is the whole point: frequent revisit is what turns a picture into monitoring. The nominal pair changes as satellites age — Sentinel-2C took over from Sentinel-2A in January 2025, and 2A has since flown an extended campaign adding acquisitions — so check which satellites contributed to the dates you use, since each unit’s spectral response differs slightly. But the effective cadence is always lower than the orbital one, because clouds remove dates and only the clear-sky observations actually contribute. Planning around Sentinel-2 therefore means thinking in terms of how many usable, cloud-free looks you expect over your area and season, not the nominal revisit alone.

Quality concerns and how they shape choices

Because it is an optical sensor, Sentinel-2’s honest limitations are atmospheric. Clouds hide the surface outright, cloud shadows darken pixels next to clouds and mimic real change, thin haze and cirrus subtly depress and distort reflectance, and there are the ordinary scene- and tile-edge effects of a tiled, orbit-based acquisition. None of these are defects to complain about; they are the conditions you design around. The first design decision is the processing level. Imagery is available both as top-of-atmosphere reflectance and as an atmospherically corrected surface-reflectance product, and for any comparison across dates or places surface reflectance is what makes the numbers mean the same thing (the Digital Imagery concept explains why). Processing changes matter too. Since processing baseline 04.00 (25 January 2022), products carry an additive offset of −1000 in their stored values, so reflectance is computed differently before and after that date; mixing dates without applying the offset — or without using a harmonized collection, or the Collection-1 reprocessing that brings the whole archive to one recent baseline — injects a false step into a time series.

The second decision is masking. Every date should pass through a cloud and shadow mask before you compute anything, following the Cloud Masking concept; Sentinel-2 ships quality and scene-classification information intended for exactly this, and there are well-established masking approaches built on it. The third decision is compositing. Because no single date is fully clear over a large area, a common and powerful pattern is to combine many masked dates into a cloud-free composite — a seasonal median, for example — that fills gaps and suppresses residual noise while preserving the signal you care about. Reflectance level, masking, and compositing are not separate chores; together they are how you turn a stack of imperfect optical scenes into a comparable measurement.

Where Sentinel-2 fits across the workflows

Most of this site’s applied workflows lean on Sentinel-2, and seeing why ties the data source back to the analysis. Its red, near-infrared, and red-edge bands make it a natural fit for vegetation work, so it underpins the NDVI Monitoring workflow and the field-scale reasoning in Agricultural Analysis, where frequent revisit and fine detail let you track individual parcels through a season. Its spatial detail and consistent archive support mapping the built environment over time, which is the basis of the Urban Expansion workflow, and its shortwave-infrared and near-infrared bands, where water absorbs almost everything, make surface-water extent and change legible — the kind of multi-date comparison the Change Detection concept formalizes (what is in the water is a separate question, covered by Optical Water Quality). In each case the sensor is chosen because its bands, detail, and cadence match the question, not because it is the only option: for broader or longer historical context it is often paired with a coarser or older record, and the sibling Landsat page covers the complementary long-archive source.

Accessing the data

Sentinel-2 data are free and open. The official distributor is the Copernicus Data Space Ecosystem, which replaced the older Copernicus Open Access Hub in 2023 and offers STAC and other catalog APIs alongside browser and processing tools. The imagery is also mirrored in several cloud-hosted archives described by the same STAC metadata; Python code and custom pipelines usually read from those, fetching just the bands, tiles, and dates a job uses, and Google Earth Engine exposes it as ready-to-query image collections for trying the masking-and-compositing pattern without downloading anything. Every route involves the same choices: which scenes for which place and dates, a cloud-cover filter, a reflectance level and processing baseline, and a mask.

Sources